OpenAI releases GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for low-latency voice agents with reasoning and tool use
Key Questions
What are the new GPT-Realtime models released by OpenAI?
OpenAI released GPT-Realtime-2.1 and GPT-Realtime-2.1-mini, which add reasoning and tool use to low-latency voice agents. The mini version maintains low costs while introducing thinking-before-speaking for better production use.
How much faster are the new Realtime models?
Latency is reduced by 25% through caching improvements. This makes them more practical for developers building real-time voice interfaces in the API.
Are the new GPT-Realtime models open source?
No, the models are commercial releases available only through OpenAI's API. They focus on concrete advancements in voice agent capabilities rather than open-source access.
OpenAI dropped two new Realtime voice models with reasoning and tool use. The mini version keeps costs low while adding thinking-before-speaking—huge for production voice agents. Latency cut by 25% via caching. This is a practical update for developers building voice interfaces. Not open-source, but a flagship commercial release that fits the focus on concrete model announcements.